A Multiplex Graph-based Recommendation System for Multi-Criteria Profiles Enhancement

Multiplex graph manipulation

By Foutse Yuehgoh, my former PhD student co-advised with Sonia Djebali at ESILV in CIFRE at Coexel. [theses.fr, manuscrit]

A real pleasure to directed Foutse those years, a long time chating and discussing. Memorable moments!

The exponential growth of the Internet has increased the availability of data, making traditional methods of finding relevant information more difficult and highlighting the need for innovative recommendation systems that utilise network science. As big data brings new challenges to understanding complex, multi-dimensional user interactions in recommender systems, some solutions use graphs such as multi-layer or multiplex graphs to represent the intricate relationships between users and items. However, these approaches often lack flexibility and adaptability in dynamic environments. To address these challenges, we present a novel framework that leverages a graph database to compute and manipulate multiplex graphs tailored to capture different user interactions. Our approach provides flexible dimension management and increases expressive power through a special algebra designed for accurate and dynamic modelling of user preferences. This algebra ensures that all necessary operators are available to manipulate the multiplex graph within the database, enabling the creation of comprehensive recommendation types. We rigorously compare our method with traditional matrix methods, including Random Walk with Restart (RWR), RWR with Singular Value Decomposition (SVD) and Tensor Factorisation. The results show that our multiplex graph approach not only provides deeper insights into users’ preferences by integrating ratings from different graph layers, but also outperforms matrix-based methods in different configurations. The strength of our framework lies in its ability to develop sophisticated, customised recommendation queries that improve performance and explainability and ultimately give users control over their data. This study represents a significant step in the development of multidimensional recommendation strategies and lays the foundation for more advanced, user-centred models that meet the urgent need for effective tools in the age of big data.

Laisser un commentaire